Bruce, Thanks for the advice. I've run through the list, redone processing and started the analysis with said list in mind, and the most plausible solution is that I shouldn't be using a crystalline material for the FEFF calculation. One interesting thing (at least to my untrained eye) to note is that the shapes of the first path and the peak that I'm trying to fit (this is before any attempt at fitting) are nearly identical, but the FEFF path is shifted about 2~3 angstroms higher than the peak. The only improvement that I could muster is that the e0 variable has gone down from 80 to 50. -Dave Baker ------------------------------------------------------------------------ -- North Carolina State University Dept. of Physics 2700 Stinton Dr. Box 8202 Raleigh, NC 27695 Ph: 919-515-5017 Fax:919-515-7331 email: dabaker@unity.ncsu.edu ------------------------------------------------------------------------ -- On Tuesday 28 October 2003 11:00 pm, Dave Baker wrote:
I am a novice student in the EXAFS world, and I am starting to work my way through Athena and Artemis. Currently, I'm working with amorphous GeTe and the Atoms/FEFF runs are based on the crystalline data. I have performed a fit in Artemis of the Ge K edge, and my e0 for three contributing paths is exceedingly large, (-70 ~ -80 volts). Two questions: 1.) What symptom is this an indicator of, and 2.) What are some possible solutions?
Hi Dave, Glad to see you were able to get started using the software after that very short conversation we had at X11A a couple weeks ago. I have seen this effect on occassion when, for some silly reason, I have made the Feff calculation using the wrong edge (i.e. K instead of L3 or vice versa). Another possible indication of an absurdly large e0 shift might be if the distances between atoms in the feff calculation are very different than the distances between atoms in the real material. That is, if the material used in the feff calculation is a very poor approximation of the real material. Another possibility that occurs to me from the way you phrased your question is that you are allowing e0 to float independently for each path. That's almost certainly not physically supportable. At least as a first stab at theproblem, you should use the same e0 for each path. One thing you might try is to not allow everything to vary in the fit. Try fixing e0 to be something sensible -- like 0.0. Vary all the rest of the parameters to get a sense of how the parameters behave and how well they describe the data. This sort of manual examination of the parameter space might help give you a clue of why you are getting such an odd result. HTH, B
Dave, On Wed, 29 Oct 2003, Dave Baker wrote:
Bruce, Thanks for the advice. I've run through the list, redone processing and started the analysis with said list in mind, and the most plausible solution is that I shouldn't be using a crystalline material for the FEFF calculation. One interesting thing (at least to my untrained eye) to note is that the shapes of the first path and the peak that I'm trying to fit (this is before any attempt at fitting) are nearly identical, but the FEFF path is shifted about 2~3 angstroms higher than the peak. The only improvement that I could muster is that the e0 variable has gone down from 80 to 50. -Dave Baker
E0 shifts greater than 10eV often mean that the fit is causing the model to "jump a period". This often means that the backscattering atom is incorrect or that the distance for the Feff calculation is way, way off. Peak shapes of |chi(R)| are difficult to interpret. But being 2 or 3 Angstroms off is a very serious problem. But a Feff model based on a crystal structure can usually do a fine job for amorphous systems.... I'd recommend looking at the simplest model (ie, 1 Feff path, no fitting parameters at all) in k-space. That is, from Artemis (or SixPack or Ifeffit), make a model with the first path and compare that to the data chi(k) and don't vary anything in the "fit". Doing this , the amplitudes may be off significantly -- you might put in a fixed value for sigma2 (say, 0.005Ang^2 as a rough guess, but you may want to play with that number) to get the amplitudes close, but don't worry too much about it being exact. But now, if the oscillations are way off (off in frequency or shifted), then the mode is not very good and you'll never get a good fit: either the distance is way off or the species of the backscattering atom is not right. Also if the _shape_ of the envelope around chi(k) is very different between data and model, that indicates that the backscattering atom might be wrong. Hope that helps, --Matt
participants (2)
-
Dave Baker -
Matt Newville